Explainable Graph-based Reinforcement Learning for Intrusion Detection in Cybersecurity
Arun Kumar BS, Rathnakar Achary · 2025
The rapid growth in digitalization has increased the number of cyber threats. This has challenged the traditional Intrusion Detection Systems (IDS), which depend on static rules and predefined signatures. Among the various methods available for detection, the Reinforcement Learning (RL) method is an adaptive approach that guarantees intrusion detection. Still, some of its limitations and the black-box nature weaken the trust and clearness in security-critical ecosystems. In this paper, we analyzed the need for artificial intelligence (AI) integration with intrusion detection by enabling Explainable Artificial Intelligence (XAI) techniques and Graph Neural Networks (GNNs) with an integration of RL to create an efficient and interoperable intrusion detection framework. The result is to leverage GNNs to depict complex relationships in network traffic and apply explainable AI to improve the transparency of RL models. We also analyze the issues related to key performance parameters such as scalability and model interoperability and provide the scope for future research.